Classification of iron oxide aerosols by a single particle soot photometer using supervised machine learning

Abstract. Single particle soot photometers (SP2) use laser-induced incandescence to detect aerosols on a single particle basis. Both refractory black carbon (rBC) and other light absorbing metallic aerosols, including iron oxides (FeOx), have been characterized by the SP2, but single particles cannot be unambiguously identified from their incandescent peak height (a function of particle mass) and color ratio (a measure of blackbody temperature) alone. Machine learning offers a promising 5 approach for improving the classification of these aerosols. Here we explore the advantages and limitations of classifying single

particle signals obtained with the SP2 using a supervised learning algorithm. Laboratory samples of different aerosols that incandesce in the SP2 (fullerene soot, mineral dust, volcanic ash, coal fly ash, Fe2O3, and Fe3O4) were used to train a random forest algorithm. The trained algorithm was then applied to test data sets of laboratory samples and atmospheric aerosols. This method provides a systematic approach for classifying incandescent aerosols by providing a score, or conditional probability, 10 that a particle is likely to belong to a particular aerosol class (rBC, FeOx, etc.) given its observed single-particle features. We consider two alternative approaches for identifying aerosols in mixed populations: one with specific class labels for each species sampled, and one with three broader classes for aerosols with similar properties. While the specific class approach performs well for rBC and Fe3O4 (≥99% of these aerosols are correctly identified), its classification of other aerosol types is significantly worse (only 47-66% of other particles are correctly identified). Using the broader class approach, we find a 15 classification accuracy of 99% for FeOx samples measured in the laboratory. The method allows for classification of FeOx as anthropogenic or dust-like for aerosols with effective spherical diameters from 170 to >1200 nm. The misidentification of both dust-like aerosols and rBC as anthropogenic FeOx is small, with < 3% of the dust-like aerosols and < 0.1% of rBC misidentified as FeOx for the broader class case. When applying this method to atmospheric observations taken in Boulder, CO, a clear mode consistent with FeOx was observed, distinct from dust-like aerosols.

Introduction

The single particle soot photometer (SP2) has been used over the past decade to quantify refractory black carbon (rBC) mass and internal mixing on a single particle basis (Stephens et al., 2003; Schwarz et al., 2006). Recently, the SP2 has been increasingly used to quantify other light absorbing refractory aerosols (e.g. Moteki et al. (2017); Liu et al. (2018)). In particular, 5 observations in source regions have shown that iron oxide containing aerosols from anthropogenic origins are present in the atmosphere (Liati et al., 2015; Dall’Osto et al., 2016; Adachi et al., 2016; Li et al., 2017), and these aerosols can be detected via laser-induced incandescence with an SP2 (Yoshida et al., 2016; Moteki et al., 2017). These aerosols have been found to be mostly pure iron oxides that are fractal aggregates of ∼100 nm spheroids, internally mixed (heterogeneously) with nitrogen or sulfate (Dall’Osto et al., 2016; Adachi et al., 2016; Li et al., 2017). They have been linked to transportation sources (engine 10 exhaust, traffic brake wear) and industrial sources such as steel processing (Ohata et al., 2018). Iron oxide aerosols quantified by the SP2 were referred to as FeOx in past literature (e.g. Moteki et al. (2017)), and we continue this convention here. In general, FeOx as quantified by the SP2 in atmospheric observations in past studies potentially included aerosols from both anthropogenic and non-anthropogenic sources. The mass mixing ratio and size distribution of FeOx has been quantified in East Asia, where observations suggested these aerosols were mainly from anthropogenic sources, and were also observed to 15 be significantly more prevalent than previously believed (Yoshida et al., 2016; Moteki et al., 2017; Ohata et al., 2018; Yoshida et al., 2018). These measurements have important implications for the climatic effects associated with these aerosols: the direct radiative climate effects of anthropogenic FeOx may be as important as brown carbon in some regions (Moteki et al., 2017;

Matsui et al., 2018), and modeling studies based on these measurements indicate these aerosols could also be an important source of particulate iron for the oceanic biogeochemical cycle (Matsui et al., 2018; Ito et al., 2018). 20 Improving the detection of iron oxide aerosols linked to anthropogenic sources is key to understanding their potential impact on the climate. The SP2 offers a promising method for real-time quantification of these aerosols, as previous detection

techniques are limited to off-line methods such as X-ray spectrometry (Adachi et al., 2016). However, while laser-induced incandescence can be used to quantify the mass of pure magnetite (Fe3O4) and to a lesser extent, hematite (Fe2O3) and wüstite (FeO) (Yoshida et al., 2016), the interpretation of ambient SP2 observations has been limited by the misclassification of other 25 aerosols as FeOx, including both rBC and aerosols containing metallic components from non-anthropogenic sources.

To first order, FeOx can be differentiated from refractory black carbon (rBC) because of differences between the blackbody temperature and peak incandescent signal (relative to the particle mass) associated with single particles incandescing in the laser of the SP2. However, the temperature and incandescent peak height alone are not sufficient to unambiguously identify FeOx. Because FeOx has a higher mass to incandescent signal relationship than rBC (Yoshida et al., 2016), and is generally 30 significantly rarer than rBC in the atmosphere in a source region by a factor of ∼250x (Moteki et al., 2017), the misclassification of even a small fraction of rBC as FeOx can bias the retrieved mass mixing ratio.